Set up the revised UCLA scale for loneliness

[ld-archer]

The accepted metric for loneliness seems to be the revised UCLA scale for loneliness (wiki link to original here). The revised scale was originally published by Russell et al (1980) and is scored as follows:

This scale has then been applied to the HRS and ELSA by selecting 3 questions that are available in the surveys:

• r*complac - How often R feels lack of companionship
• r*leftout - How often R feels left out
• r*isolate - How often R feels isolated from others

Hughes et al. (2004) applied this scale to the HRS and provided evidence for its efficacy, and Shankar et al. (2011) applied the same revised scaled to ELSA.

Each question is answered as ['1 - Hardly ever of never', '2 - Some of the time', '3 - Often'], which is then combined into a summary score from 3-9, with higher scores indicating greater loneliness. It's not completely clear from the paper whether Shankar et al. used the mean score of the three questions (which then means we could use the summary variable r*lnlys3) or the sum of each variable, as they mention both in the methods section. The summary score does seem to have some advantages as it handles some of the missing values (i.e. if respondent has 1 or 2 missing, they still have a score), so I'm leaning towards that right now.

ToDo:

• [x] Add each of the 4 variables (3 individual and summary score var) to the FEM
• [x] Include the loneliness var(s) in the prediction of chronic disease and mortality
  • [x] Find some literature linking loneliness to chronic disease to indicate which diseases loneliness can affect
  • [ ] Find literature linking loneliness to other risk behaviours
• [ ] #111
  • Couldn't figure out how to do this in either ELSA_vars.txt or inside the CPP code, need to talk to Bryan about this. Use the continuous atotb value as predictor for now but come back to it
• [x] Do a full test run of the model with the new loneliness variable and check the output t-tests
• [x] Create a new overleaf document and add a methods section for loneliness
  • [x] Table on deriving the revised UCLA scale (questions and scores)
  • [x] Rehash the text above into a bit of an intro/explanation of the scale, including the literature
• [x] Figure out why in some cases, the lnly variable is being assigned a value of 0 whilst running the model
  • Have a look at other ordinal vars like adlstat or exstat as I don't think these have the same issue

[ld-archer]

Set up revised UCLA scale for loneliness

[ld-archer]

Literature linking loneliness and chron disease / mortality

Loneliness, Social Isolation, and Chronic Disease Outcomes

Christiansen at al. (2020)

Found independent association between loneliness, social isolation, and both Cardiovascular disease and type 2 Diabetes.

Does loneliness contribute to mild cognitive impairment and dementia? A systematic review and meta-analysis of longitudinal studies

Lara et al. (2019)

Meta analysis showing positive association between loneliness and increased risk of Dementia, as well as limited evidence suggesting a potential effect of loneliness on Mild Cognitive Impairment (MCI).

Loneliness and Risk of Dementia

Sutin et al. (2018)

"Cox proportional hazards regression indicated that loneliness was associated with a 40% increased risk of dementia. This association held controlling for social isolation, and clinical, behavioral, and genetic risk factors. The association was similar across gender, race, ethnicity, education, and genetic risk."

Study used data from the HRS.

Social isolation, loneliness, and all-cause mortality in older men and women

Steptoe et al. (2013)

Steptoe is a prolific author in this field and generally uses ELSA in his analyses. Finds differences between health links of loneliness and social isolation.

Social Isolation

• Lung disease
• Arthritis
• Impaired mobility
• Depressive symptoms

  Loneliness

• CHD
• Stroke
• Clinical Depression

This study also found a significant link between both social isolation, loneliness, and mortality.

[ld-archer]

Outcomes currently predicted by loneliness (may change)

• Mortality
• CVD
• Diabetes
• Stroke
• Alzheimer's / Dementia
• Stroke
• High Cholesterol
• ADL/IADLs

Maybe?

• Smoking
• Alcohol consumption
• Physical Activity
• BMI (covered by physical activity?? Missing effect on diet if not included)

Need to find direct link in literature for the maybe pile. Slightly confusing as some papers say only social isolation has links to these things, some suggest both loneliness and social isolation have links. Also think about this for high cholesterol and hypertension, probably need to move into the maybe pile for loneliness (but not social isolation).

[ld-archer]

Predictors of Loneliness

• age
• gender
• wealth quintile
• education level
• nssec?
• marital status? (can't include in prediction of social isolation because this is part of the index but maybe include in loneliness?)
• work status?

[ld-archer]

Full run complete and t-tests generated. Things to note:

• Demographic prediction still good (gender, race, average age etc.)
• Of health conditions, some bad predictions including alzhe/dementia, heart disease, hypertension and lung disease. These are either directly predicted by loneliness (hearte, alzhe/demene) or strongly related (lung disease - limiting long standing illness related to social isolation and loneliness)
• Income and wealth v poor prediction, but haven't done anything with these in a while so will pick that back up soon. Will be using wealth_quint for prediction of others which will hopefully be much more robust
• Other risk behaviours look really good. See how alcohol frequency stuff looks when thats included too

[ld-archer]

Just found that in some cases, lnly is being assigned a value of 0 (seen in the testing_crossvalidation.dta testing dataset in analysis/techdoc_ELSA/). This is obviously wrong so I need to figure out why and stop it happening. I'm confident this is happening during the running of the model because we don't have any 0 values in the input populations. Adding to the checklist at the top of this issue.

[ld-archer]

Lets look at the distribution of lnly in ELSA_long.dta

Lnly

. tab lnly, m

 Loneliness |
 Score, Low |
to High [1, |
         3] |      Freq.     Percent        Cum.
------------+-----------------------------------
          1 |     47,627       47.85       47.85
          2 |     19,518       19.61       67.46
          3 |      3,050        3.06       70.52
          . |     13,320       13.38       83.90
         .c |     15,385       15.46       99.36
         .m |        585        0.59       99.95
         .r |         52        0.05      100.00
------------+-----------------------------------
      Total |     99,537      100.00

Lnly by gender

. tab lnly male,column

+-------------------+
| Key               |
|-------------------|
|     frequency     |
| column percentage |
+-------------------+

Loneliness |
Score, Low |
   to High |         Male
    [1, 3] |         0          1 |     Total
-----------+----------------------+----------
         1 |    25,599     22,028 |    47,627 
           |     64.93      71.59 |     67.85 
-----------+----------------------+----------
         2 |    11,769      7,749 |    19,518 
           |     29.85      25.18 |     27.81 
-----------+----------------------+----------
         3 |     2,058        992 |     3,050 
           |      5.22       3.22 |      4.35 
-----------+----------------------+----------
     Total |    39,426     30,769 |    70,195 
           |    100.00     100.00 |    100.00 

Looks like women are slightly more likely to score higher on the loneliness scale, however the vast majority of people rank their loneliness as low.

Lnly by Educ

. tab lnly educ, column

+-------------------+
| Key               |
|-------------------|
|     frequency     |
| column percentage |
+-------------------+

Loneliness |
Score, Low |
   to High |    Harmonised Education Level
    [1, 3] | 1.Less th  2.Upper S  3.Tertiar |     Total
-----------+---------------------------------+----------
         1 |    12,562     22,472      8,702 |    43,736 
           |     61.79      69.79      74.10 |     68.05 
-----------+---------------------------------+----------
         2 |     6,454      8,515      2,772 |    17,741 
           |     31.75      26.44      23.61 |     27.60 
-----------+---------------------------------+----------
         3 |     1,314      1,213        269 |     2,796 
           |      6.46       3.77       2.29 |      4.35 
-----------+---------------------------------+----------
     Total |    20,330     32,200     11,743 |    64,273 
           |    100.00     100.00     100.00 |    100.00 

People with a lower education level are more likely to have a high loneliness score (3). Probably what we would have expected, but the same is not true for social isolation, which is pretty interesting.